One of the most common mistakes in data analysis is choosing a chart type before thinking about what question you're trying to answer. The chart comes last, not first. Before you open any charting tool, ask: what relationship am I trying to show?
There are five fundamental relationships that charts are built to communicate. Everything else is a variation of one of these.
Comparison: how do values differ across categories?
Bar chart — The workhorse of comparison. Each category gets a bar; the bar height represents the value. Use vertical bars when your categories have short names; horizontal bars when names are long or there are many categories (easier to read left-to-right). Never use a pie chart for comparisons between more than two items.
Grouped bar chart — When you want to compare sub-categories within each main category. Useful but gets cluttered beyond three or four sub-groups.
Column chart with reference line — A bar chart with a horizontal line showing a target, average, or prior period. Immediately shows which categories are above or below the benchmark.
Trend: how does a value change over time?
Line chart — The default for time series. The connected line emphasises continuity and direction. Works well for a small number of series (three to four) plotted together.
Area chart — A filled version of the line chart. Emphasises volume rather than just direction. Stacked area charts show how a total is composed over time, but can be misleading when the top series is hard to read.
Candlestick / OHLC chart — Specialised for financial data, showing open, high, low, and close values for each period. Not appropriate for general data.
Distribution: how are values spread?
Histogram — Groups continuous values into bins and shows how many observations fall into each bin. Essential for understanding the shape of a distribution before doing any statistical analysis. Not the same as a bar chart (bars represent counts of ranges, not categories).
Box plot (box-and-whisker) — Shows the five-number summary: minimum, Q1, median, Q3, maximum, and outliers. Extremely efficient for comparing distributions between groups. One box plot communicates more statistical information than most charts three times its size.
Violin plot — A box plot with the distribution shape overlaid. Useful when the shape matters, such as when data is bimodal.
Relationship: how do two variables relate to each other?
Scatter plot — Each point represents one observation, with x and y coordinates for two variables. The pattern of dots reveals correlation, clusters, non-linear relationships, and outliers. Add a trend line to show the direction of the relationship. One of the most powerful charts in the toolkit.
Bubble chart — A scatter plot where each point has a size representing a third variable. Use carefully — size is hard to compare accurately, so this works best when the third variable represents something proportional like population or revenue.
Heatmap — Shows a matrix of values as a colour grid. Excellent for correlation matrices, weekly patterns (weekday × hour), and any situation where you have two categorical axes and one numerical value.
Part-to-whole: how does each part contribute to the total?
Pie chart — Use only when you have two to five categories and the total must add up to 100%. Pie charts are frequently misused for comparisons (bar charts do it better) or for too many segments (becomes unreadable). The human eye is poor at comparing angles.
Donut chart — A pie chart with a hole in the middle. Slightly easier to read because the arc lengths are clearer than full pie slices. Often used for KPI displays where a single percentage is shown in the centre.
Treemap — Shows hierarchical data as nested rectangles, sized by value. Useful for portfolio analysis, disk usage, budget breakdowns. Difficult to read at small sizes.
Waterfall chart — Shows a running total as sequential additions and subtractions. Classic for profit-and-loss statements and explaining how you moved from one period's value to another.
Three rules that cover most situations
Rule 1: Match the chart to the relationship, not the other way around. If you find yourself forcing data into a pie chart because "it looks nice," stop. Pick the chart that makes the relationship easiest to read.
Rule 2: Fewer is usually better. A chart with two series is almost always more readable than one with six. If you have many series, consider small multiples (the same chart repeated for each sub-group) rather than piling everything into one panel.
Rule 3: Label clearly or not at all. A chart with an unclear title, unlabelled axes, and no units is worse than a table. Every chart needs a title that answers the "so what?" question, not just "what is plotted."